Ex vivo 18F-fluoride uptake and hydroxyapatite deposition in human coronary atherosclerosis

Early microcalcification is a feature of coronary plaques with an increased propensity to rupture and to cause acute coronary syndromes. In this ex vivo imaging study of coronary artery specimens, the non-invasive imaging radiotracer, 18F-fluoride, was highly selective for hydroxyapatite deposition in atherosclerotic coronary plaque. Specifically, coronary 18F-fluoride uptake had a high signal to noise ratio compared with surrounding myocardium that makes it feasible to identify coronary mineralisation activity. Areas of 18F-fluoride uptake are associated with osteopontin, an inflammation-associated glycophosphoprotein that mediates tissue mineralisation, and Runt-related transcription factor 2, a nuclear protein involved in osteoblastic differentiation. These results suggest that 18F-fluoride is a non-invasive imaging biomarker of active coronary atherosclerotic mineralisation.

Coronary atherosclerosis is an inflammatory disease that results in the formation of intimal plaque with an increased propensity to rupture. Microscopic calcification is a key feature of ruptured atherosclerotic plaques and the identification of coronary microcalcification is closely linked to coronary thrombotic events 1 . However, the in vivo mechanisms governing the accumulation of early microscopic calcification within the coronary vasculature are poorly understood. Pre-clinical studies have proposed atherosclerotic inflammation to be an initiator of plaque calcification through the extrusion and response to calcifying extracellular vesicles 2,3 . Additionally, in vitro models of intimal plaque microcalcification have demonstrated that spherical or ellipsoidal micro-calcifying vesicles aggregate within plaques and coalesce to form larger plates of macrocalcification 3 . Whilst the transition from microcalcification to macrocalcification in the vast majority of plaques is thought to confer stability, the presence of micro-calcifying vesicles in the tunica intima has the potential to reduce the structural integrity of thin-capped fibroatheroma, resulting in plaque rupture 4,5 .
Recently, studies have demonstrated that increased 18 F-sodium fluoride ( 18 F-fluoride) positron emission tomography (PET) uptake is observed in culprit plaques following myocardial infarction and in plaques with multiple adverse features in patients with stable disease 1,6-8 . 18 F-Fluoride preferentially binds to exposed hydroxyl groups on the surface of nanocystalline hydroxyapatite. We have previously demonstrated that the signal intensity of 18 F-fluoride in carotid endarterectomy specimens increases as the size of the calcifications decrease, such that 18 F-fluoride is an imaging biomarker of unbound microscopic calcification 9,10 . However, there are important differences between carotid and coronary atherosclerotic plaque progression, predominantly attributed to plaque composition. Compared with carotid plaques, vulnerable coronary plaques are more prone to rupture owing to thinner fibrous caps (< 65 µm versus < 200 µm) and a reduction in smooth muscle cells in the tunica media 11 . To address these differences in pathophysiology and to fully understand the mechanisms of 18 F-fluoride binding in coronary atherosclerotic plaque, direct histological examination of coronary artery tissue is warranted 12 . In this study, we performed an ex vivo histological validation of 18 F-fluoride binding to calcium derivatives and osteogenic proteins involved in human coronary atherosclerotic calcification. www.nature.com/scientificreports/ pericardium from thoracic aortic dissection and two deaths were related to non-cardiac causes (suffocation and alcohol toxicity). None of the coronary artery specimens represented culprit or ruptured plaque directly related to the sudden death.
18 F-Fluoride co-localisation with hydroxyapatite. Within tissue sections from the 13 patients, 32 coronary artery plaques were identified using high-resolution micro-computed tomography. Specifically, the morphology and architecture combined with the CT attenuation number of the tissue differentiated coronary artery plaques (CT number > 300) from the surrounding myocardium (CT number < 100) (Fig. 1C). 18 F-Fluoride binding was observed in plaques both with (n = 19) and without (n = 13) areas of macroscopic calcification as determined on micro-CT. Total plaque 18 F-fluoride binding (median 157.5 [IQR 103.9-216.9] kBq/mL) was more than tenfold greater than non-specific binding in the surrounding myocardium (median 14.9 [IQR 9.6-27.4] kBq/mL, p < 0.0001) and more than 500-fold greater than background regions (median 0.3 [IQR 0.09-1.3] kBq/mL, p < 0.0001) (Fig. 1D). 18  In plaques with no observable macrocalcification on micro-computed tomography, nanocrystalline calcification was detected using high-intensity 18 F-fluoride in focal regions co-localised to fibroatheromatous plaques ( Fig. 2A-D). In plaques with macrocalcification, high-intensity 18 F-fluoride uptake was observed in distinct regions remote from larger macrocalcific deposits identified on micro-computed tomography (Fig. 2F,G,K,L). Additional low-intensity 18 F-fluoride uptake was observed on the exposed surface of macrocalcification (Fig. 2G,L). In the high-intensity regions, 18   As there was an asymmetrical peak to signal in these macrocalcified specimens, further analysis of the area under the whitlockite peak revealed two overlapping peaks attributable to hydroxyapatite and whitlockite at a ratio of 30:70.

F-Fluoride co-localisation with markers of osteogenic activity in coronary arteries. Detailed
analysis of coronary artery specimens revealed that 18 F-fluoride binding was predominantly observed within the tunica intima in areas of plaque formation (Fig. 4A). Little or no binding was observed in the tunica media.

Discussion
In this ex vivo imaging study of coronary atherosclerosis, we have demonstrated for the first time that 18 F-fluoride is a selective marker of intimal hydroxyapatite deposition in human coronary atherosclerotic plaques. Similar to other disease states it preferentially binds in areas of coronary microcalcification rather than macrocalcification. 18 F-Fluoride has a high affinity for hydroxyapatite, which has a higher surface area for binding in regions of microcalcification compared with larger macrocalcified deposits. Importantly, high 18 F-fluoride signal colocalises with the distribution of osteopontin and Runx-2, established markers of early calcification activity and adverse plaque formation. This histological validation supports the use of 18 F-fluoride positron emission tomography as a marker of developing microcalcification and plaque activity in patients with coronary artery disease. Whilst there has been histological confirmation of 18 F-fluoride binding in carotid atheroma, studies demonstrating increased 18 F-fluoride activity in the coronary arteries 1,7 have been called into question due to the limited spatial resolution of clinical positron emission tomography, with some investigators questioning whether 18 F-fluoride binding actually occurs in coronary arteries 14 . In this regard, the confirmation of high-intensity 18 F-fluoride binding in the intimal layer of coronary plaques compared to background and adjacent myocardium is of considerable importance. We have also confirmed that 18 F-fluoride binding occurs in plaques both with www.nature.com/scientificreports/ and without macroscopic calcium observed on CT, and that binding appears to occur preferentially in regions of developing microcalcification. These findings are consistent with previous observations in coronary artery disease as well the data in carotid atheroma and other cardiovascular disease states 9,10,15-17 .
Of particular interest is the potential for 18 F-fluoride to discriminate hydroxyapatite deposition above other calcium derivatives in regions of active mineralisation 18 . Of the many calcium derivatives, nanocrystalline hydroxyapatite is the central component of microcalcification in atherosclerotic coronary plaques 18 . We have here confirmed the preferential binding of 18 F-fluoride for microcalcification and for hydroxyapatite based upon the binding of a specific optical probe and Raman spectroscopy. At later stages in the calcification process, other calcium derivatives, such as whitlockite, become more abundant in calcified vascular tissue, particularly within large vessel atherosclerosis where there is often a high whitlockite to hydroxyapatite ratio 19 . The phase transformation of hydroxyapatite to whitlockite may occur in the hypoxic or acidic conditions within necrotic cores where magnesium ions are incorporated onto the surface and prevent further growth of hydroxyapatite crystals 20 . Traditionally surface area effects have been used to explain the preferential binding of 18 F-fluoride for microcalcification. However, the specificity of 18 F-fluoride for hydroxyapatite provides an additional explanation for why high 18 F-fluoride activity is not observed in areas macrocalcification and why it provides different information to CT 9 .
We also observed a close relationship between the coronary 18 F-fluoride signal and both osteopontin and Runx-2 expression, established markers of early calcification activity and adverse coronary plaque. High concentrations of osteopontin accumulate in coronary atheroma exposed to hypoxia and endothelial injury 21 . Of note, inflammatory signalling within metabolically active coronary plaques stimulates macrophage-derived foam cells to express high levels of osteopontin 22 . In comparison, low levels of osteopontin mRNA are found in vascular smooth muscle cells often regarded as the cell type responsible for initiating plaque calcification 23 . Importantly, from a clinical perspective, high plasma osteopontin levels are associated with adverse clinical events in patients with both stable and unstable coronary artery disease 24,25 . Combined with high-sensitivity C reactive protein, osteopontin had a two-fold increased risk of recurrent myocardial infarction in patients presenting with ST elevation myocardial infarction 25 . The role of osteopontin in mediating plaque activity is noted by the beneficial effect of statins in reducing osteopontin levels and thereby reduce the risk of plaque rupture 26 . The relationship between 18 F-fluoride uptake and osteopontin therefore supports its role as a marker of early calcification activity and adverse plaque formation. However, ultimately data are required to investigate whether 18 F-fluoride predicts future myocardial infarction and therefore might provide important clinical information. In this regard, to determine whether coronary 18 F-fluoride has clinical utility, prognostic observation studies in patients with recent myocardial infarction are ongoing (NCT02278211).
Co-localisation of 18 F-fluoride uptake to specific coronary atheromatous plaques presents some challenges when conducting clinical positron emission tomography and computed tomography coronary angiography in patients with coronary artery disease. The small calibre of the coronary arteries leads to partial volume averaging, and the near continuous motion from cardiac and respiratory cycles can limit signal localisation to specific regions of the coronary circulation. Some of these issues can be improved by use of beta-blockade, motion correction and advanced image analysis techniques 27 . The present study is therefore important to reaffirm that 18 F-fluoride is binding to individual advanced human coronary atheromatous plaques as well as identify the components to which it binds. The emerging clinical application of 18 F-fluoride positron emission tomography and computed tomography coronary angiography is able to provide clinicians with a tool to monitor disease activity and identify individuals at increased risk of future coronary events 28 . Since the probability of an individual ruptured plaque causing an acute coronary event is low, 18 F-fluoride positron emission tomography and computed tomography coronary angiography is best utilised to detect overall coronary atherosclerotic disease activity in vulnerable patients rather than localising specifically to a single vulnerable plaque 29 .
There are some limitations to this study. Legislation regarding the regulation of tissue in victims of sudden death meant that only left main and proximal left anterior descending coronary artery specimens could be obtained for detailed research analysis. These specimens did not include sections of culprit coronary plaque rupture with thrombus formation and therefore extrapolation of these findings to ruptured atherosclerotic plaques cannot be made. Additionally, specimen preparation and handling between different imaging modalities may have resulted in alterations in the orientation and alignment of the datasets. Ante-mortem demographics regarding risk factors such as diabetes mellitus and renal disease which influence atherosclerotic calcification were unavailable. However, the majority of cases in this study were adjudicated by a forensic pathologist who determined a cause of death attributed to ischemic heart disease independent from the study investigators. This provides further evidence of the high prevalence of 18 F-fluoride binding in coronary arteries in a high-risk cohort. Although the significance of the high frequency of sudden cardiac death (77%) in this study population is uncertain, further studies exploring the utility of coronary 18 F-fluoride imaging in victims of sudden death are worth pursuing.

Conclusions
In this ex vivo study of coronary atherosclerotic plaques, 18 F-fluoride binding was highly selective for unbound hydroxyapatite deposition. High 18 F-fluoride intensity was associated with intimal microcalcification in regions of osteopontin and Runx-2 expression. This study provides further evidence to support the use of 18  www.nature.com/scientificreports/

Methods
Saturation binding assays to quantify 18 F-fluoride binding kinetics and selectivity to hydroxyapatite. Saturation radioligand binding experiments to determine the number of binding sites (Bmax) and the dissociation constant (K d ) of 18 F-fluoride were undertaken using nanocrystalline hydroxyapatite phantoms prior to performing ex vivo imaging. Five-milligram vials of hydroxyapatite were incubated with 18 F-fluoride (110, 230, 470 or 700 kBq/mL) for 20 min. The supernatant fraction was then removed and hydroxyapatite was twice washed in 10 mL 0.9% sodium chloride for 5 min to remove unbound 18 F-fluoride. Hydroxyapatite phantoms were scanned using high-resolution micro-PET (1:5 coincidence mode) and computed tomography (CT) with semi-circular full trajectory, maximum field of view, 480 projections, 50 kVp, 300 ms and 1:4 binning (Mediso nanoScan PET/CT, Mediso Medical Imaging Systems, Hungary) and total activity counts over 30 min were measured. PET data were reconstructed using Mediso's iterative Tera-Tomo 3D reconstruction algorithm using 4 iterations, 6 subsets, full detector model, normal regularization, spike filter on, voxel size 0-6 mm and 400-600 keV energy window. Micro-PET-CT images were analysed on an OsiriX workstation (OsiriX version 7.5.1, 64-bit, OsiriX Imaging Software, Geneva, Switzerland). Regions of interest were drawn around contours of phantoms on the CT and mapped to corresponding fused 18 F-fluoride positron emission tomographic images. Total binding activity curves and Scatchard plots were generated to calculate Bmax and K d of 18 F-fluoride for subsequent ex vivo experiments. To ensure saturation of binding sites, 2 × K d was used to evaluate the selectivity of 18 F-fluoride for hydroxyapatite compared with phantoms of calcium phosphate, calcium oxalate and calcium pyrophosphate using the method described above. . Specimens were twice washed in 10 mL 0.9% sodium chloride for 5 min to remove unbound 18 F-fluoride. Specimens were scanned using the microPET-CT protocol described above. Regions of interest were drawn in background regions, myocardium, non-calcified and calcified segments in coronary artery plaques using micro-CT images and the maximum 18 F-fluoride activity in each region was recorded on co-registered micro-PET images. Maximum activity recorded in a region equal to or above 100 kBq/mL was defined as high-intensity 18 F-fluoride (> 5 × myocardium activity), whereas values with a maximum activity of less than 100 kBq/mL was defined as low-intensity. After whole specimen imaging, the coronary arteries were fixed in 10% (w/v) neutral buffered formalin.

Sample preparation and histological examination.
Formalin-fixed coronary arteries were sectioned into 2-4 mm slices by a vascular biology/pathology lab (MS, SS). The resulting slices were embedded in paraffin which provided 1-8 tissue cross-sections analysed per sample depending on the size of tissue initially isolated. Paraffin sections (4 µm) were used for histology and immunohistochemistry as described below. In all cases, images were generated using an Aperio Slide Scanner using ImageScope software (Leica Biosystems, Germany). Histological examination was performed using haematoxylin and eosin staining for overall pathology, followed by Movat's pentachrome and trichome staining to differentiate fibrosis and elastic fibres. Von Kossa and Alizarin Red S staining were used to assess for the presence of calcification.
Fluorescein-bisphosphonate immunofluorescence of cadaveric coronary arteries. To determine whether the binding of 18 F-fluoride in regions of Von Kossa and Alizarin Red S was specific for hydroxyapatite deposition, categorisation of these regions using a fluorescein-bisphosphonate probe was undertaken. Fluorescein-bisphosphonate is a highly sensitive and specific probe for identifying regions of microcrystalline hydroxyapatite. Incubation and binding in tissue has previously been described in detail 13 . Briefly, sections were de-waxed in xylene and incubated with fluorescein-bisphosphonate (1 µM) for 2 h, washed in water (2 ×) followed by incubation with 2% Alizarin Red S (250 µL) for 5 min. Sections were washed in water (3 ×) and subsequently incubated with 4′,6-diamidino-2-phenylindole (500 nM) for 5 min. Sections were washed with water (1 ×) and then mounted using ProLong Gold Antifade. Fluorescence signal was detected under a Leica DMRB fluorescence microscope.
Raman spectroscopy. Coronary artery specimens were embedded in paraffin wax, sectioned and subsequently placed on calcium fluoride slides. Once dried, the slides were placed in xylene for 15 min followed by dehydration in ethanol. As soon as they were dehydrated, the sections were ready for imaging and did not require additional processing. Alizarin Red S staining was used to discriminate regions of calcification. Sections from tissue presenting no calcification, calcification with low fluoride intensity (< 100 kBq/mL), and no calcification with high fluoride intensity (≥ 100 kBq/mL) were selected in order to address whether there were any differences in the apatite crystal or molecular substitution in the structure. Raman imaging was carried out using an InVia Renishaw Microscope with a 785 nm laser excitation source which was used to excite the sample through a 50, N.A. 0.75 objective. The total data acquisition was performed during 60 s for spectra with a 100% www.nature.com/scientificreports/ laser power using the WiRE software (Renishaw, Gloucestershire, United Kingdom). All of the spectra acquired were background subtracted using a background correction algorithm.  DS9390). The omission of the primary antibody served as negative controls. Blinded qualitative categorisation of nuclear staining intensity was performed by a pathologist using a 4-point classification: 0, no notable staining, 1, < 20% of plaque or relevant cells are weakly positive, 2, < 50% of plaque or some relevant cells are strongly positively, 3, 50-100% of plaque or all relevant cells are strongly positive. For comparison of immunohistochemical analysis with 18 F-fluoride activity, plaques with a classification 3 were defined as 'positive' and classification < 3 was defined as 'negative' for osteopontin, runtrelated transcription factor 2, transforming growth factor beta 1 and wingless/integrated 3a. Plaques with classification 2 were defined as 'positive' and classification < 2 were defined as 'negative' for Caspase 3.

Statistical analysis.
Categorical variables are reported as number (%) and continuous variables as mean ± standard deviation for parametric or median and interquartile range for non-parametric data. Normality was tested for using the D' Agostino and Pearson test. Continuous unpaired variables were compared using Student's t test with Welch's correction when two samples had unequal variances and/or unequal sample sizes. Non-parametric data was compared between two categories using Mann-Whitney U test or using Kruskal-Wallis test for multiple categories. Statistical analysis was undertaken using PRISM for OS X, version 8.